The desert robot shows great potential for AI

As a young man, Hongzhi Gao lived with his family in Gansu, a region in northwestern China near the Tengger Desert. Reflecting on his childhood, he recalls that outside his house there was a steady and steady storm of mud, and that for many months of the year he stepped out of the house before the sand filled the empty space and entered his house. bags, shoes, and mouth. Desertification in the desert remained in his head for many years, and at the university he turned the memorial into an idea of ​​designing a machine that could bring plants to the desert.

Efforts to control the spread of desertification — the way in which fertile soil becomes a wilderness — have prioritized expensive handicrafts. Hongzhi designed a robot with in-depth learning skills to change the way planting trees: from identifying the right place to planting tree seedlings to watering. Although unfamiliar with AI, as a undergraduate student, Hongzhi used the in-depth Baidu PaddlePaddle platform to stitch together the various components to make a robot more capable of recognizing objects than similar machines already in the market. It took less than a year for Hongzhi and his associates to finalize the final version and use it.

The Hongzhi desert robot is a perfect example of an ingenious invention.

Today, more than four million developers are using Baidu’s open AI technology to create solutions that can improve the lives of people in their communities, and many of them have little or no skills in the field. “In the next ten years, AI will be a source of change in our region, transforming the way industries and businesses operate. AI developer.

As we enter a new chapter on AI transformation, Haifeng Wang, CTO of Baidu, identified two key factors that support the company’s approach: AI will continue to mature and increase its challenges. And at the same time, the cost of shipping and entry restrictions will be reduced – benefiting all businesses that build AI solutions on a large scale and developers exploring the AI ​​world.

Combining knowledge with data and in-depth learning

The combination of knowledge and data and in-depth learning has greatly contributed to the functionality and accuracy of AI models. Since 2011, Baidu’s AI architecture has been discovering and integrating new features into larger information platforms. Currently, this information contains more than 550 billion facts, which cover all aspects of daily life, as well as topics related to industry, including manufacturing, medicine, law, economic activity, technology, media and entertainment.

This knowledge and key points all together form the building blocks of the new Baidu-trained PCL-BAIDU Wenxin model (version ERINIE 3.0 Titan). This model exceeds the examples of other languages ​​without the graphs of knowledge on 60 natural language (NLP) tasks, including reading comprehension, writing, and semantic similarity.

Points in different ways

Extensive learning is a new phase of AI research that seeks to improve machine understanding and better mimic human change. Examples of research efforts in this area include the integration of words and images, where the model is taught to create images from self-explanatory descriptions, as well as algorithms designed to comprehend visual and visual comprehension of words. The challenge with these functions is for machines to create semantic connections on a variety of dataset (e.g., images, text) and to understand the interdependence between them.

The result of AI is the integration of AI technologies such as computer vision, speech recognition, and the adaptation of natural languages ​​to create multi-purpose machines.

In the future, Baidu unveiled its own NLP genre that combines language and emotional comprehension. Examples of specific applications of this type include digital avatars that can detect human-like environments and support clients in companies, and algorithms that are able to “draw” pieces of art and write poetry based on their understanding of architectural design. .

There are also many creative, efficient products that can be developed in this field. The PaddlePaddle platform is able to create intellectual and linguistic connections, which enabled a group of Chinese masters students to develop a dictionary to preserve endangered languages ​​in places like Yunnan and Guangxi for easy translation into simple Chinese.

Integration of AI into software and hardware, as well as corporate applications

While AI systems are used to solve a growing number of industry-related problems, greater emphasis is placed on the integration of software (in-depth learning framework) and hardware (AI chip) as a whole, rather than individual performance, taking into account factors such as computing power, use performance power, as well as latency.

In addition, major innovations are taking place on Baidu’s AI infrastructure platform, where third-party developers are using advanced learning capabilities to create new applications based on specific applications. The PaddlePaddle platform contains a number of APIs that support the use of AI in the latest technologies such as quantum computing, life science, computational fluid mechanics, and cell power.

AI also has useful functions. For example, in Shouguang, a small town in Shandong province, AI is used to set up a fruit and vegetable market. It only takes two people and one program to manage multiple web pages.

And this is obvious, says Wang, “Although the number of AI technologies is increasing, an open platform for in-depth learning brings processor and applications as operating systems, reducing barriers to entry for companies and individuals who want to integrate AI into their business.”

Reducing the entry barriers for manufacturers and end users

In the forefront of technology, already training major models like PCL-BAIDU Wenxin (version of ERNIE 3.0 Titan) has overcome many of the barriers that face traditional cultures. For example, these multi-objective models have helped lay the groundwork for managing a wide range of sub-NLP tasks, such as distributing notes and answering questions, in a single integrated environment, where in the past, each type of task should be eliminated. is a different model.

PaddlePaddle also has a number of tools that help developers, such as integrated color technologies to change the colors of all objects to suit specific scenarios. The platform provides a comprehensively supported library of multinational corporations with over 400 species, from large to small, which have only a small fraction of all national sizes but are capable of achieving uniform functionality, reducing model development and shipping costs.

Today, Baidu’s open-source technology expertise supports a team of more than 4 million AI creators who created 476,000 colors, enabling the transformation of 157,000 businesses and organizations under AI-led design. The examples cited above are the results of the latest developments in all aspects of Baidu AI architecture, which integrate technologies such as voice recognition, computer vision, AR / VR, graphical graphs, and already taught major examples that are one step closer to awareness. the world as a people.

Instead, AI has reached a critical level that allows it to perform amazing functions. For example, the recent launch of the Metaverse XiRang would not have been possible without the PaddlePaddle platform to create digital avatars so that participants around the world could connect to their devices. In addition, future breakthroughs in places like quantum computing could improve the performance of metaverses. This shows how Baidu’s various offerings are intertwined and interdependent.

In a few years, AI will be closer to the core of our human experience. It will be for our people how steam, electricity, and the internet were for generations past. As AI becomes more complex, developers like Hongzhi will be working harder as architects and architects, given the freedom to explore previously used cases that were considered feasible. Heaven is the limit.

This was created by Baidu. Not written by MIT Technology Review authors.

Leave a Reply

Your email address will not be published. Required fields are marked *